DIRECTIONAL SPEECH RECOGNITION DEVICE AND ETHOD
Technical Field
The invention relates to the field of speech recognition.
Background
There is an emergent market for voice-controlled multi- modal, multimedia and telematic devices. This market raises the problem that such devices must be able to discern whether or not an utterance was addressed to that particular device, or to some third party.
With devices that use voiced keywords or voiced dialling, one does not want the device to perform actions as a consequence of overhearing chance instances of those keywords or names whilst it rests on a desk or in a pocket. The consequences may be for the device to inadvertently call someone mentioned in an overheard conversation, or to change mode/application, making it confusing to the user to find the device interface changing apparently ^randomly' from use to use. Both behaviours would be seen as a severe disadvantage to voice control by the user.
One cannot simply rely on the volume level to differentiate between overheard casual speech and close-talking use as with a normal telephone, because multimedia and multi-modal devices and telematic control systems are generally intended to be used at arms length, either to view a display or because the device is on the dashboard.
Thus there is a need for an alternative method.
US 6219645 Bl (Byers) and JP 2002091491 A (Sanyo) both describe placing multiple microphones in fixed positions in a room, enabling localisation of a user with the intent of associating commands then uttered with devices distributed within the room. However, such methods are not applicable to mobile telephony, for example.
US 5884254 A (Ucar) suggests the use of separate microphone arrangements for speech transmission and speech recognition functions respectively, but this would appear to incur unnecessary cost, complexity and weight.
The general approach in the art is to use a non-obvious keyword to precede interaction with the device, for example giving it a name to be addressed by. However, having to call your personal appliances by name is not likely to be a practical or particularly desirable solution in many circumstances .
Summary of the Invention
In a first aspect, the present invention provides a speech recognition device, as claimed in claim 1.
In a second aspect, the present invention provides a method for controlling speech recognition, as claimed in claim 11.
Further aspects are as claimed in the dependent claims .
Brief description of the drawings
FIG. 1 illustrates a possible microphone configuration in accordance with a preferred embodiment of the invention;
FIG. 2 illustrates a target direction for the user's voice, together with an angular threshold of deviation theta from the target direction.
FIG. 3 illustrates an arc of possible target directions for the user's voice, ranging between 'normal to the plane of the user interface' and 'parallel to the plane of the user interface' .
FIG. illustrates a system for the control of speech recognition in accordance with the preferred embodiment.
Detailed description of preferred embodiment
In a preferred embodiment, a speech recognition device is described in accordance with figure 4, comprising an input signal 404 and an estimate of the direction 406 of the input signal. The target direction selected from the target options stored in store 412 is compared with the direction 406 of the input signal by a processor 430. This comparison uses a threshold of deviation from the target direction stored in 410 to determine whether the input signal 404 should be passed as output 408 to a recogniser.
Figure 1 illustrates the preferred embodiment, comprising at least three microphones 160, 162, 164 that provide input to a means to determine the direction of a signal source. The speech recognition process is then controlled, by: (I) Comparing the direction of the signal source with a given target direction. Target directions are illustrated as direction 210 of figure 2, and directions generally shown as 314 in figure 3; and
(n) Permitting recognition processing if the source direction is within an angular threshold 220, 320 of the target direction.
FIG. 2 illustrates a target direction for the user's voice that is substantially normal to the plane of the user interface and microphones, together with an angular threshold of deviation theta from the target direction
In figure 2, the initial target direction 210 assumes that the user will wish to look at the user interface 250 of device 200 when controlling it. Consequently in this preferred embodiment the microphones should be distributed to form a plane substantially in parallel with the plane of the user interface 250. The target direction 210 can then
be taken to be normal to the plane of the microphones and thus implicitly the plane of user interface 250. It is anticipated that the user will rarely be aligned exactly normal to the plane of the user interface 250, and so an angular threshold 220 is introduced, wherein the speech recognition process is still permitted if the user is within angle θ 230 of the target direction.
An alternative target direction can be selected either by the user via a user interface 450 or by automatic control 440 if the device is placed in a power and/or data cradle, or if it is left alone on a substantially horizontal surface. This target direction is described in accordance with figure 3.
FIG. 3 illustrates an arc of possible target directions for the user's voice, ranging between 'normal to the plane of the user interface' and 'parallel to the plane of the user interface' , in the direction downwards and vertical to the plane of the user interface, together with an angular threshold of deviation theta from the possible target directions .
So the alternative target direction may be anywhere on an arc 314 that is centred on the vertical axis of the plane of the user interface 350. Arc 314 extends between the normal to the plane of the user interface 310, and procedes down the vertical axis until parallel to the plane of the user interface 312.
It is anticipated that the user need not be exactly aligned on this arc. Therefore an angular threshold 320 is introduced, wherein the speech recognition process is still
permitted if the user is within angle θ 330 of the target direction .
The angular control of the speech recognition system can be specified by the user in situations where the user needs to control the device from a relative position other than those permitted by the configurations above. In such an instance the user may manually indicate the wish to redefine the target direction via the user interface 450, and then speak from that direction to set the device.
In the preferred embodiment, the user's direction is determined by comparing the relative signal delay between pairs of microphones 160, 162, 164, and then using these delays and the positions of the corresponding microphones to calculate the signal direction.
The angular control of the speech recognition system can be overridden automatically in situations where the signal from any one microphone 160, 162, 164 falls below an amplitude ratio with respect to the signals from the remaining microphones, as in when the device is held in a typical phone position, so favouring reception by any microphone near the mouth.
Finally, in the preferred embodiment it is desirable to provide a facility to reversibly enable or disable the angular control of the speech recognition system via the user interface 450, for example in a dictation scenario.